Three Way Multi Model Interoperation

1
Three Way Multi Model Interoperation

• Lee W.Wagenhalsltlwagenhagt_at_gmu.eduAdversary
  Behavioral Modeling
• Maxwell AFB, Montgomery AL
• March 18 - 19, 2008

2
Abstract

• A process and interoperation technique for using
  three modeling tools (ORA, CASEAR III, and
  Pythia) has been demonstrated using the East
  Africa Embassy Bombing (1998) as an example
• Outline
• Approach
• Models
• Results

3
Approach

• With respect to the suite of models available
  from CASOS, CSC, and SAL the question of how data
  or information can be passed between the models
  is unknown or un-proven.
• A Limited Discovery Experiment was used to
  explore the potential interoperation between
  modeling techniques to determine if
• interoperation is possible,
• various interoperation types can be applied
• use of such interoperation would improve the
  overall analysis over that provided by the models
  independently.
• A case study approach was taken using a corpus of
  data about the al Qaeda Bombings of the Embassy
  in Kenya.

4
Approach

• A single data source, the Anatomy of a Terrorist
  Attack, was used as the input for the
  Demonstration

Controls ?

• The information in the document was fed into the
  CASOS tool via Automap and edited by the analysts
  who used read the data set

Analyst
Social Network Analysis with Meta Matrix
Data Set About Situation, e.g. East Africa
Embassy Bombing
ORA
Al Qaeda cell organization structure
(Interconnections, minos and maxos

• Outputs from ORA were used to create CAESAR III
  and Pythia models

Probability profiles for COA analysis

• Analytical products were produced from the three
  tools

5
Types of Interoperation

• The basic proposition Understanding of an
  adversary and analysis of Effects based Courses
  of Action can be improved by using these multiple
  models by exchanging information or data between
  them.
• Three types of interoperation have been
  postulated
• Human to human (Swivel Chair) (Human gains
  insights from model 1 that helps with the human
  set up or analysis of model 2)
• Data to Data (files can be exported from one
  model that can be read automatically by another
  model under the direction of the analyst)
• Automated Model to Model (Model can be connected
  over a network and automatically exchange data as
  they are run)

6
Experiment (Interoperation) Process Example

• 1. Analyst sets up Automap and Runs ORA
  generating Meta Matrix and XML File for CAESAR 3
  and Pythia
• 2a. CAESAR III analyst uses ORA file plus
  knowledge from Data Set to generate adversary
  organization models (lattice plus CP net).
  Identifies potential communications links for ISR
• 2b. The analyst loads ORA file into Pythia and
  refines the model using ORA Meta Matrix in Pythia
• 3. Analyst uses TIN to produce probability
  profiles, comparing COAs for selection.
• 4. COA is selected, planned, and executed
  including ISR Tasking
• 5. Indicator data from ISR used to update Data
  Set and Pythia for on-going Assessment

5 Indicator data
Analyst
1. Meta Matrix Attributes
Data Set About Situation, e.g. East Africa
Embassy Bombing
1. Adversary Structure, Organization, Dynamics
Meta Matrix
ORA
2b. File (Agents/ Roles/Task/ Location)
2a. File (Agents/ Roles/Task))
4. Plan Selection and Execution
2a. Adversary Organization Dynamics
3. Intervention COA Probability Profiles
5 Indicator data
7
Step 1 ORA SN (agent to agent view) for East
Africa Embassy Bombing
Case 1
Case 1

• Opened new Pythia and re-ran the model Case 1
  with low marginals, Now we get the expected
  results

8
ORA Agents to Knowledge
9
Step 1 ORA SN (agent to Task view) for East
Africa Embassy Bombing
10
The Kenya Team
The following command structure was inferred from
reports (text) and ORA meta matrix
11
The Kenya Team

• The same representation in CAESAR III

12
All Structures

• Lattice Algorithm reveals all possible
  organizational structures (28) of the Al Qaeda
  Kenya cell

13
Minimum Interactions Between Cell Members
Command
Output to the environment Bombing
14
Adversary Model of Plan(Based On SNA of Kenya
Attack)
Model was created using Human to Human
interoperation

• Derived from events and their timing as described
  in Anatomy
• The actors and their roles from the SNA and
  CAESAR 3 map to the events
• h and g values assigned based on understanding of
  TIN

15
Adversary Model of Plan w/Blue Interventions

• Hypothetical Blue actions (as might be perceived
  by the adversary) added

16
Model Created From ORA Analysis
17
Demonstrate Potential

• The tools and the workflow indicate that
  interoperation can provide a synergist set of
  useful results

18
Problem Definition

• Given
• Multiple Knowledge Representations
• Multiple Reasoning/Computational Approaches
• Modeling/Analysis Objective
• A set of questions to be answered by the analyses
  performed on the computational models
• Solution
• Identification of Model(s) and/or Combinations of
  Models that offer insight into the solution space
• Workflow

19
Problem Definition

• Nexus Between Models
• What query can be generated in one model that can
  be answered by the other?
• What are the overlaps among the models?
• How do we determine if the output of one is
  supported by the output of the other?
• How do we identify gaps, inconsistencies, or
  incompleteness (need for more information)?
• Workflow
• Given an analysis objective, what is the workflow
  (i.e., combination, interactions, and sequence
  of/between models) that exploits the
  multi-modeling nexus in addressing the objective?

20
Setup Model View
Modeling/Analysis Objective
21
Theoretical Challenge
?
Rule Models Ontologies
22
Types of Results Achieved

• Better Model Construction by
• Providing design parameters for the construction
• e.g., Social Network to Organization structure.
• Providing the structure (partially and/or
  completely)
• e.g., Social Network to Timed Influence Net.
• Model Validation
• Results from two models support each other
• Multi-agent model and Social Network simulation
  models.
• Enhancements to analysis capabilities of a model
  by employing functionality from another
• e.g., Temporal analysis of Timed Influence Net
  modes.
• Construction of new models by embedding multiple
  models in a single framework
• e.g., Organization and Communication models.

23
Conclusion

• Three way interoperation between models has been
  demonstrated
• One tool and its model can assist in the creation
  of a different model in another tool
• Used a combination of human swivel chair and
  data-to-data interoperation
• Process workflows are being developed.
• More effort needed to refine workflows and
  interoperation techniques
• Need to extend the approach to more models